changelog

What's new in
CodeScape.

Track every feature, improvement, and fix. We ship fast and we ship often.

v2.1.0 April 2026

Any AI. Any CLI. Any Language.

latest

The release that breaks every remaining lock-in. A custom endpoint slot turns the provider list infinite. A native C++ MCP bridge plugs CodeScape into Claude Code, Cursor, Codex and Gemini CLI without a single Python dependency. And because your AI provider already speaks every major human language, prompt CodeScape in whichever one you think in. The responses come back in kind. Plus the Templates Library, Rule Packs, smarter prompt rewriting, the Zhipu GLM provider, and a level-design fix that stops the "everything is 100x too big" failure mode.

Provider & CLI Reach

  • Custom OpenAI-compatible endpoint: alongside the 11 preconfigured providers, you can now point CodeScape at any endpoint that speaks the OpenAI chat/completions spec. That covers OpenRouter, Together, Fireworks, Perplexity, vLLM, LM Studio, llama.cpp, internal company gateways, and every future LLM that ships with an OpenAI-style API. New model drops next month? Plug in its URL, no plugin update required.
  • Native C++ MCP bridge: the in-editor MCP server (JSON-RPC 2.0 over localhost) is now a first-class C++ subsystem with zero Python dependencies. Claude Code, Cursor, Codex, Gemini CLI, and any other MCP-compatible client can drive the full 321-tool surface from outside the editor. Bound to 127.0.0.1 only, OFF by default.
  • Multilingual prompting, native: CodeScape passes your prompt through to the AI provider verbatim. Since every modern LLM is natively multilingual, you can write prompts in English, French, Spanish, German, Japanese, Chinese, Portuguese, Arabic, Korean (whatever you think in) and the model replies in the same language. No translation layer, no setup, no extra tokens burned on a wrapper. Works with all 11 preconfigured providers and any custom endpoint.

Beginner-Friendly Prompting

  • Templates Library : 10 curated multi-pass recipes accessible from a new ✦ Templates button. Click Gothic Church Greybox, Health & Damage System, or Magic Portal and the prompt is dropped into your input fully structured, with explicit passes and the right tool names. Works with any provider, but recommends the best fit per template via a non-blocking toast.
  • Rule Packs : 6 stackable presets in Global AI Rules: Strict Greybox, Rapid Prototype, Production Quality, Multiplayer-Ready, C++ First, and Plan First (Safety Net). Toggle them on a per-project basis. Each pack is capped at ~125 tokens so combining 3-4 still leaves your context window healthy.
  • Prompt Reformulator v2 : the wand button (Ctrl+Shift+R) now rewrites drafts using CodeScape-specific best practices: numbered PASSES, explicit tool names, size: in cm, anti-pattern callouts. One-shot call, never pollutes chat history.
  • Zhipu GLM provider (z.ai) : 11th native provider, OpenAI-compatible. Two menu entries: Standard (pay-per-token) and Coding Plan ($18/mo flat for heavy coding). Models: glm-5.1, glm-5v-turbo, glm-4-plus, glm-4-air, glm-z1-air. Up to 200K context on glm-5.1.
  • size: parameter on spawn_actors : target dimensions in centimeters with auto-scale derivation from mesh bounds. Fixes the most common level-design footgun: passing scale:{x:3000} on a 100 cm Cube no longer produces a 300-meter monster.

Quality of Life

  • System prompt WORKFLOW F (Level Design) gains a SIZING section that drills the cm-vs-multiplier distinction into every model.
  • Templates and rule packs are data-driven via Resources/DefaultTemplates.json and Resources/DefaultRulePacks.json. Drop your own overrides into Saved/CodeScape/user_templates.json or user_rule_packs.json : same schema, no recompile.
  • Atomic file writes for the rule-pack toggle state : no corruption if the editor crashes mid-toggle.
  • Game-thread guards on the new singletons : catches misuse early in dev builds, silent in shipping.
  • Mistral default model bumped from mistral-small-latest to mistral-medium-latest. Small's 32K context is too tight for CodeScape's system prompt; the menu now warns about it explicitly.

Fixes

  • Square-character bug in the chat header (UTF-8 byte escapes inside TEXT() macros), replaced with proper \\uXXXX Unicode escapes.
  • JSON parsing edge case in template/rule-pack loaders: malformed entries are now skipped with a Verbose log instead of failing the whole file.
  • Append-only enum for EApiProvider::GLM, existing providers.json files keep their indices intact (no silent provider type shifts).
  • Cleaner em-dash-free copy across the website and product, less "AI-written" tone.

Branding

  • New CodeScape logo (isometric blue cube inside a hexagon) deployed everywhere: plugin Icon128.png, website favicon, nav, footer, page headers, and team card.
  • Bigger, more visible logo presence across descriptive sections without being intrusive.
v2.0.0 March 2026

The Complete Engine

The release that reframed what an AI plugin for Unreal Engine could be. 321 tools, 79 categories, 175K+ lines of native C++, no Python, no external runtime. CodeScape stops being a chat overlay and becomes a full game-creation engine driven by natural language.

What Ships

  • 321 tools across 79 categories, the industry's largest AI toolkit for UE5. Spawn actors, wire Blueprints, author materials, generate Niagara systems, build cinematics, sculpt landscapes, configure Lumen and Nanite, all from natural language.
  • 10 AI providers at launch: Gemini, OpenAI, Claude, DeepSeek, MiniMax, Groq, Mistral, Cohere, Grok, plus Ollama for fully-local inference. Bring your own keys, no markup, no middleman.
  • NodeForge Blueprint Compiler: a 13-phase DSL-to-graph layout engine that generates production-ready node graphs with proper pin routing, reroute nodes, and comment containment, no manual cleanup needed.
  • Native C++ MCP Server: a JSON-RPC 2.0 bridge so Claude Desktop, Cursor, and any MCP-capable client can drive the full 321-tool surface from outside the editor.
  • Per-asset contextual sessions: docked AI panels auto-injected into Blueprint, Material, Widget, AnimBP, and BehaviorTree editors. The AI sees what you're editing without you having to describe it.

Tool Coverage

  • Niagara VFX: particle systems, fire/smoke/rain presets, emitter and module configuration.
  • Sequencer & Cinematics: 14 tools covering level sequences, camera rigs, keyframes, focal-length tracks.
  • Audio & MetaSounds: ambient placement, spatial audio, sound mixes, procedural MetaSound patches.
  • Landscape: terrain creation, sculpting, painting, heightmap import/export.
  • PCG Framework: graph editing, presets, spatial filters and sampling nodes.
  • World Building: procedural structure types (walls, towers, arches, staircases, houses, mansions, castles, towns).
  • Chaos Physics: destruction, vehicles, cloth, rigid bodies, constraints.
  • Advanced Rendering: Lumen, Nanite, Virtual Shadow Maps, TSR, Ray Tracing, Pixel Streaming.
  • AI & Gameplay: behavior trees, AI perception, NavMesh, EQS, Gameplay Abilities, Enhanced Input.
  • Advanced Systems: Control Rigs, State Trees, Smart Objects, Mass Entity, Groom (hair), Water.
  • Localization: string scanning, string tables, translation pipeline.
  • Asset Management: find, rename, validate, audit, dependency tracking.

Architecture

  • 100% native C++20, zero external dependencies, no Python bridges.
  • 175,000+ lines of production code across 793 files.
  • O(1) tool dispatch via TMap registry.
  • FScopedTransaction undo support on 151+ tools.
  • Automatic pre-modification asset backups, every destructive operation can be rolled back.
  • Streaming response support for all cloud providers.
  • Dynamic 40KB+ system prompt generation, context-aware injection of relevant tool definitions.
  • Activity logging and usage analytics subsystems for telemetry-free local diagnostics.

Quality

  • 22 regression test suites covering Blueprints, Materials, Niagara, Sequencer, Audio, AI, and end-to-end pipelines.
  • Comprehensive audit tracking and bug reporting infrastructure.
  • Thread-safe static map initialization.
  • 30-second operation timeout for deadlock prevention on long-running tools.
v1.0.0 2025

Foundation

The first public release. Proves the thesis: a native C++ AI plugin can drive Unreal Engine 5 from natural language without Python, without remote bridges, and without compromising on responsiveness.

Core

  • Module-driven plugin architecture with proper lifecycle management.
  • Streaming AI session widget docked in the editor.
  • Multi-provider support at launch: Gemini, OpenAI, Claude, DeepSeek.
  • Tool registry with O(1) dispatch, designed to scale to hundreds of tools.
  • First tool families: Blueprint creation/editing, Material authoring, level design and actor management.
  • Project Settings integration for API keys and behavior toggles.

Stay up to date.

Follow us on Discord and GitHub for the latest updates and announcements.

Join Discord Follow on GitHub